Build failure in blead: Linux -Dusequadmath -Dcc=g++-8 #16689
Comments
From @jkeenanI first caught a whiff of this build-time failure in this smoke-test http://perl.develop-help.com/raw/?id=213970 ... whose log file is here: http://perl5.test-smoke.org/logfile/70257 (Note the 6 "Segmentation fault" messages in that log.) I installed gcc-8 and g++-8 locally on Ubuntu 18.04 LTS and configured thus: ##### (The perl_V.txt attached does not pertain to this build.) make test_prep failed after miniperl had been built, apparently during Same configuration only with gcc-8 tests PASS. Can someone diagnose? Thank you very much. |
From @jkeenan./miniperl -Ilib make_ext.pl lib/auto/POSIX/POSIX.so MAKE="make" LIBPERL_A=libperl.a LINKTYPE=dynamic |
From @jkeenanSummary of my perl5 (revision 5 version 29 subversion 3) configuration: Characteristics of this binary (from libperl): |
From @sisyphusOn Tue, 11 Sep 2018 15:24:21 -0700, jkeenan@pobox.com wrote:
FWIW, I get the same failures on Ubuntu 18.04 LTS using g++ version 7.3.0. Cheers, |
|
The RT System itself - Status changed from 'new' to 'open' |
From @tonycozOn Tue, 11 Sep 2018 15:24:21 -0700, jkeenan@pobox.com wrote:
This wasn't g++ 8 (or 7) specific - I reproduced it with g++ 6 too. The problem is the C++ standard provides three versions of log10(): log10(float) but we're calling it with a different floating point type: log10(__float128) so the compiler doesn't know which to call. As with many other math functions, perl.h defines Perl_log10, which becomes log10q on quadmath builds, changing POSIX.xs to use Perl_log10 fixes the problem. Similarly for ldexp(), except that using Perl_ldexp() takes a slightly larger change. With those changes compilation succeeded, but two tests for signbit() failed. Since quadmath provides signbitq() I changed perl.h to use that for quadmath builds. Tony |
From @jkeenanOn Wed, 12 Sep 2018 01:36:56 GMT, tonyc wrote:
Tony, thanks for getting right on this. So far looks good on g++-8. I'm taking this ticket for the purpose of monitoring smoke tests for a few days, after which I'll close the ticket. Thank you very much. |
From @jkeenanOn Wed, 12 Sep 2018 02:15:42 GMT, jkeenan wrote:
Unfortunately, we're getting some new smoke test failures in one particular configuration. ##### Observed in these smoke test reports: 1. 2. 3. Branch in each case was blead or a branch rebased on blead after e6081c0 -- |
From @jkeenanOn Wed, 12 Sep 2018 19:26:31 GMT, jkeenan wrote:
Here is the failure: ##### ok 1 - No perlio debug file without -Di... Test Summary Report run/switchDx.t (Wstat: 0 Tests: 8 Failed: 4) Configuration attached. |
From @jkeenanSummary of my perl5 (revision 5 version 29 subversion 3) configuration: Characteristics of this binary (from libperl): |
From @tonycozOn Wed, 12 Sep 2018 12:39:55 -0700, jkeenan wrote:
The attached should fix it. Tony |
From @tonycoz0001-perl-133510-can-t-use-my_vsnprintf-on-quadmath-threa.patchFrom 14b21593c19501eb30d65a4572dce272c08f12cd Mon Sep 17 00:00:00 2001
From: Tony Cook <tony@develop-help.com>
Date: Thu, 13 Sep 2018 11:45:14 +1000
Subject: (perl #133510) can't use my_vsnprintf() on quadmath/threaded builds
Since anything with quadmath should be recent enough to have
vsnprintf() I've fallen back to that.
Calls to PerlIO_debug() in core don't include floating point values
and I expect it to be unlikely outside of core - if it is needed
they'll just have to use double formatting/types.
---
perlio.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/perlio.c b/perlio.c
index f5eb4851b6..3a2f9120d3 100644
--- a/perlio.c
+++ b/perlio.c
@@ -371,7 +371,19 @@ PerlIO_debug(const char *fmt, ...)
/* Use fixed buffer as sv_catpvf etc. needs SVs */
char buffer[1024];
const STRLEN len1 = my_snprintf(buffer, sizeof(buffer), "%.40s:%" IVdf " ", s ? s : "(none)", (IV) CopLINE(PL_curcop));
+# ifdef USE_QUADMATH
+# ifdef HAS_VSNPRINTF
+ /* my_vsnprintf() isn't available with quadmath, but the native vsnprintf()
+ should be, otherwise the system isn't likely to support quadmath.
+ Nothing should be calling PerlIO_debug() with floating point anyway.
+ */
+ const STRLEN len2 = vsnprintf(buffer + len1, sizeof(buffer) - len1, fmt, ap);
+# else
+ STATIC_ASSERT_STMT(0);
+# endif
+# else
const STRLEN len2 = my_vsnprintf(buffer + len1, sizeof(buffer) - len1, fmt, ap);
+# endif
PERL_UNUSED_RESULT(PerlLIO_write(PL_perlio_debug_fd, buffer, len1 + len2));
#else
const char *s = CopFILE(PL_curcop);
--
2.11.0
|
From @jkeenanOn Thu, 13 Sep 2018 01:49:15 GMT, tonyc wrote:
It works for me with that configuration on both g++7 and g++8. So I'm pushing it to blead to let the smokers have another go at it. Thank you very much. -- |
From @jkeenanOn Thu, 13 Sep 2018 02:27:07 GMT, jkeenan wrote:
Smoke tests look good. Re-closing. -- |
|
@jkeenan - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#133510 (status was 'resolved')
Searchable as RT133510$
The text was updated successfully, but these errors were encountered: